Re: [Boost-users] [EXTERNAL] Re: Boost Units initialize quantity with signaling NaN

Subject: Re: [Boost-users] [EXTERNAL] Re: Boost Units initialize quantity with signaling NaN
From: Belcourt, Kenneth (kbelco_at_[hidden])
Date: 2016-08-13 15:38:14

> On Aug 13, 2016, at 10:56 AM, Steven Watanabe <watanabesj_at_[hidden]> wrote:
> 
> AMDG
> 
> On 08/10/2016 08:55 PM, Belcourt, Kenneth wrote:
>> 
>> I’m trying to find a way to initialize a Boost Unit with a signaling NaN, not just a NaN with the signal cleared. It seems that the only way to initialize a quantity is by multiplying by the unit, and that multiplication clears the signal on the NaN.  I know I can just modify the quantity.hpp header and initialize the default constructor val_ member with the signaling NaN, but I was hoping for a cleaner solution.
>> 
>> Here’s an example (typed in by hand, not tested):
>> 
>>  const double s_nan = std::numeric_limits<double>::signaling_NaN();
>> 
>>  typedef quantity<si::acceleration, double> acceleration_t;
>>  BOOST_UNITS_STATIC_CONSTANT(acceleration_u, acceleration);
>> 
>>  acceleration_t a(s_nan * acceleration_u);  // <— this clears the signal on the NaN
>> 
>> Has anyone encountered this or know of a more elegant solution than modifying the quantity.hpp header?
>> 
> 
> I can't reproduce the problem with VS 2015.
> 
> #include <boost/units/quantity.hpp>
> #include <limits>
> #include <boost/units/systems/si.hpp>
> #include <iostream>
> 
> using boost::units::quantity;
> namespace si = boost::units::si;
> 
> int main() {
>    quantity<si::length> x(std::numeric_limits<double>::signaling_NaN()
> * si::meter);
>    std::cout << std::hex << *(unsigned long long*)&x << std::endl;
>    return 0;
> }
> 
> prints 7ff8000000000001
> 
>  Multiplication by a unit (as opposed to
> a quantity) does not require a floating
> point multiplication.  It only copies the
> value around.  I have no idea why you're
> seeing the signal bit being cleared.

I wonder if copying is somehow causing problems.  I see this:

  Copying a NaN may not preserve its bit representation.

mentioned here:

  http://en.cppreference.com/w/cpp/types/numeric_limits/signaling_NaN


Perhaps that’s our problem.

Thanks for the help Steven.